Merge pull request #401 from trustgraph-ai/release/v1.0

1.0 -> master

containers/Containerfile.base

@@ -4,14 +4,16 @@
 # like to avoid re-doing this.
 # ----------------------------------------------------------------------------
 
-FROM docker.io/fedora:40 AS base
+FROM docker.io/fedora:42 AS base
 
 ENV PIP_BREAK_SYSTEM_PACKAGES=1
 
-RUN dnf install -y python3 python3-pip python3-wheel python3-aiohttp && \
+RUN dnf install -y python3.12 && \
-    dnf clean all
+  alternatives --install /usr/bin/python python /usr/bin/python3.12 1 && \
-
+  python -m ensurepip --upgrade && \
-RUN pip3 install --no-cache-dir pulsar-client==3.5.0
+  pip3 install --no-cache-dir wheel aiohttp && \
+  pip3 install --no-cache-dir pulsar-client==3.7.0 && \
+  dnf clean all
 
 # ----------------------------------------------------------------------------
 # Build a container which contains the built Python packages.  The build

containers/Containerfile.bedrock

@@ -4,14 +4,16 @@
 # like to avoid re-doing this.
 # ----------------------------------------------------------------------------
 
-FROM docker.io/fedora:40 AS base
+FROM docker.io/fedora:42 AS base
 
 ENV PIP_BREAK_SYSTEM_PACKAGES=1
 
-RUN dnf install -y python3 python3-pip python3-wheel python3-aiohttp \
+RUN dnf install -y python3.12 && \
-    python3-rdflib
+  alternatives --install /usr/bin/python python /usr/bin/python3.12 1 && \
-
+  python -m ensurepip --upgrade && \
-RUN pip3 install --no-cache-dir boto3 pulsar-client==3.5.0
+  pip3 install --no-cache-dir wheel aiohttp && \
+  pip3 install --no-cache-dir pulsar-client==3.7.0 && \
+  dnf clean all
 
 # ----------------------------------------------------------------------------
 # Build a container which contains the built Python packages.  The build

containers/Containerfile.flow

@@ -4,21 +4,25 @@
 # like to avoid re-doing this.
 # ----------------------------------------------------------------------------
 
-FROM docker.io/fedora:40 AS base
+FROM docker.io/fedora:42 AS base
 
 ENV PIP_BREAK_SYSTEM_PACKAGES=1
 
-RUN dnf install -y python3 python3-pip python3-wheel python3-aiohttp \
+RUN dnf install -y python3.12 && \
-    python3-rdflib
+  alternatives --install /usr/bin/python python /usr/bin/python3.12 1 && \
+  python -m ensurepip --upgrade && \
+  pip3 install --no-cache-dir wheel aiohttp rdflib && \
+  pip3 install --no-cache-dir pulsar-client==3.7.0 && \
+  dnf clean all
 
 RUN pip3 install --no-cache-dir \
     anthropic cohere mistralai openai google-generativeai \
     ollama \
-    langchain==0.3.13 langchain-core==0.3.28 \
+    langchain==0.3.25 langchain-core==0.3.60 \
-    langchain-text-splitters==0.3.4 \
+    langchain-text-splitters==0.3.8 \
-    langchain-community==0.3.13 \
+    langchain-community==0.3.24 \
     pymilvus \
-    pulsar-client==3.5.0 cassandra-driver pyyaml \
+    pulsar-client==3.7.0 cassandra-driver pyyaml \
     neo4j tiktoken falkordb && \
     pip3 cache purge
 

containers/Containerfile.hf

@@ -4,22 +4,26 @@
 # like to avoid re-doing this.
 # ----------------------------------------------------------------------------
 
-FROM docker.io/fedora:40 AS ai
+FROM docker.io/fedora:42 AS ai
 
 ENV PIP_BREAK_SYSTEM_PACKAGES=1
 
-RUN dnf install -y python3 python3-pip python3-wheel python3-aiohttp \
+RUN dnf install -y python3.12 && \
-    python3-rdflib
+  alternatives --install /usr/bin/python python /usr/bin/python3.12 1 && \
+  python -m ensurepip --upgrade && \
+  pip3 install --no-cache-dir wheel aiohttp && \
+  pip3 install --no-cache-dir pulsar-client==3.7.0 && \
+  dnf clean all
 
 RUN pip3 install torch==2.5.1+cpu \
     --index-url https://download.pytorch.org/whl/cpu
 
 RUN pip3 install --no-cache-dir \
-    langchain==0.3.13 langchain-core==0.3.28 langchain-huggingface==0.1.2 \
+    langchain==0.3.25 langchain-core==0.3.60 langchain-huggingface==0.2.0 \
-    langchain-community==0.3.13 \
+    langchain-community==0.3.24 \
-    sentence-transformers==3.4.0 transformers==4.47.1 \
+    sentence-transformers==4.1.0 transformers==4.51.3 \
-    huggingface-hub==0.27.0 \
+    huggingface-hub==0.31.2 \
-    pulsar-client==3.5.0
+    pulsar-client==3.7.0
 
 # Most commonly used embeddings model, just build it into the container
 # image

containers/Containerfile.ocr

@@ -4,14 +4,16 @@
 # like to avoid re-doing this.
 # ----------------------------------------------------------------------------
 
-FROM docker.io/fedora:40 AS base
+FROM docker.io/fedora:42 AS base
 
 ENV PIP_BREAK_SYSTEM_PACKAGES=1
 
-RUN dnf install -y python3 python3-pip python3-wheel python3-aiohttp \
+RUN dnf install -y python3.12 && \
-    python3-rdflib tesseract poppler poppler-utils
+  alternatives --install /usr/bin/python python /usr/bin/python3.12 1 && \
-
+  python -m ensurepip --upgrade && \
-RUN pip3 install --no-cache-dir pytesseract pulsar-client==3.5.0
+  pip3 install --no-cache-dir wheel aiohttp && \
+  pip3 install --no-cache-dir pulsar-client==3.7.0 && \
+  dnf clean all
 
 # ----------------------------------------------------------------------------
 # Build a container which contains the built Python packages.  The build

containers/Containerfile.vertexai

@@ -4,15 +4,17 @@
 # like to avoid re-doing this.
 # ----------------------------------------------------------------------------
 
-FROM docker.io/fedora:40 AS base
+FROM docker.io/fedora:42 AS base
 
 ENV PIP_BREAK_SYSTEM_PACKAGES=1
 
-RUN dnf install -y python3 python3-pip python3-wheel python3-aiohttp \
+RUN dnf install -y python3.12 && \
-    python3-rdflib
+  alternatives --install /usr/bin/python python /usr/bin/python3.12 1 && \
-
+  python -m ensurepip --upgrade && \
-RUN pip3 install --no-cache-dir \
+  pip3 install --no-cache-dir wheel aiohttp && \
-    google-cloud-aiplatform pulsar-client==3.5.0
+  pip3 install --no-cache-dir pulsar-client==3.7.0 && \
+  pip3 install --no-cache-dir google-cloud-aiplatform && \
+  dnf clean all
 
 # ----------------------------------------------------------------------------
 # Build a container which contains the built Python packages.  The build
@@ -47,5 +49,3 @@ RUN \
 
 WORKDIR /
 
-
-

trustgraph-base/trustgraph/api/flow.py

@@ -24,7 +24,7 @@ class Flow:
         else:
             return self.api.request(f"flow", request)
 
-    def id(self, id="0000"):
+    def id(self, id="default"):
         return FlowInstance(api=self, id=id)
 
     def list_classes(self):

trustgraph-base/trustgraph/api/knowledge.py

@@ -39,7 +39,7 @@ class Knowledge:
 
         self.request(request = input)
 
-    def load_kg_core(self, id, user="trustgraph", flow="0000",
+    def load_kg_core(self, id, user="trustgraph", flow="default",
                      collection="default"):
 
         # The input consists of system and prompt strings
@@ -53,7 +53,7 @@ class Knowledge:
 
         self.request(request = input)
 
-    def unload_kg_core(self, id, user="trustgraph", flow="0000"):
+    def unload_kg_core(self, id, user="trustgraph", flow="default"):
 
         # The input consists of system and prompt strings
         input = {

trustgraph-base/trustgraph/api/library.py

@@ -208,7 +208,7 @@ class Library:
         return {}
 
     def start_processing(
-            self, id, document_id, flow="0000",
+            self, id, document_id, flow="default",
             user="trustgraph", collection="default", tags=[],
     ):
 

trustgraph-base/trustgraph/base/request_response_spec.py

@@ -127,8 +127,12 @@ class RequestResponseSpec(Spec):
 
         rr = self.impl(
             client = processor.pulsar_client,
+
+            # Make subscription names unique, so that all subscribers get
+            # to see all response messages
             subscription = (
-                processor.id + "--" + flow.name + "--" + self.request_name
+                processor.id + "--" + flow.name + "--" + self.request_name +
+                "--" + str(uuid.uuid4())
             ),
             consumer_name = flow.id,
             request_topic = definition[self.request_name],

trustgraph-bedrock/setup.py

@@ -34,7 +34,7 @@ setuptools.setup(
     python_requires='>=3.8',
     download_url = "https://github.com/trustgraph-ai/trustgraph/archive/refs/tags/v" + version + ".tar.gz",
     install_requires=[
-        "trustgraph-base>=0.23,<0.24",
+        "trustgraph-base>=1.0,<1.1",
         "pulsar-client",
         "prometheus-client",
         "boto3",

trustgraph-cli/scripts/tg-graph-to-turtle

@@ -71,8 +71,8 @@ def main():
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-invoke-agent

@@ -110,8 +110,8 @@ def main():
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-invoke-document-rag

@@ -39,8 +39,8 @@ def main():
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-invoke-graph-rag

@@ -47,8 +47,8 @@ def main():
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-invoke-llm

@@ -47,8 +47,8 @@ def main():
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     args = parser.parse_args()

trustgraph-cli/scripts/tg-invoke-prompt

@@ -42,8 +42,8 @@ def main():
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-load-doc-embeds

@@ -191,8 +191,8 @@ async def main(running):
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-load-kg-core

@@ -13,7 +13,7 @@ from trustgraph.api import Api
 import json
 
 default_url = os.getenv("TRUSTGRAPH_URL", 'http://localhost:8088/')
-default_flow = "0000"
+default_flow = "default"
 default_collection = "default"
 
 def load_kg_core(url, user, id, flow, collection):

trustgraph-cli/scripts/tg-load-pdf

@@ -86,8 +86,8 @@ def main():
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-load-sample-documents

@@ -40,8 +40,7 @@ documents = [
         "id": "https://trustgraph.ai/doc/challenger-report-vol-1",
         "title": "Report of the Presidential Commission on the Space Shuttle Challenger Accident, Volume 1",
         "comments": "The findings of the Commission regarding the circumstances surrounding the Challenger accident are reported and recommendations for corrective action are outlined",
-#        "url": "https://ntrs.nasa.gov/api/citations/19860015255/downloads/19860015255.pdf",
+        "url": "https://ntrs.nasa.gov/api/citations/19860015255/downloads/19860015255.pdf",
-        "url": "file://19860015255.pdf",
         "kind": "application/pdf",
         "date": datetime.datetime.now().date(),
         "tags": ["nasa", "safety-engineering", "space-shuttle"],
@@ -740,10 +739,3 @@ def main():
 
 main()
 
-
-
-
-# https://warwick.ac.uk/fac/soc/pais/people/aldrich/publications/beyond.pdf
-
-# https://www.ialeia.org/docs/Psychology_of_Intelligence_Analysis.pdf
-

trustgraph-cli/scripts/tg-load-text

@@ -88,8 +88,8 @@ def main():
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-load-turtle

@@ -113,8 +113,8 @@ def main():
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
     
     parser.add_argument(

trustgraph-cli/scripts/tg-save-doc-embeds

@@ -161,8 +161,8 @@ async def main(running):
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-show-flow-state

@@ -63,8 +63,8 @@ def main():
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-show-graph

@@ -39,8 +39,8 @@ def main():
 
     parser.add_argument(
         '-f', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)'
+        help=f'Flow ID (default: default)'
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-start-library-processing

@@ -54,8 +54,8 @@ def main():
 
     parser.add_argument(
         '-i', '--flow-id',
-        default="0000",
+        default="default",
-        help=f'Flow ID (default: 0000)',
+        help=f'Flow ID (default: default)',
     )
 
     parser.add_argument(

trustgraph-cli/scripts/tg-unload-kg-core

@@ -13,7 +13,7 @@ from trustgraph.api import Api
 import json
 
 default_url = os.getenv("TRUSTGRAPH_URL", 'http://localhost:8088/')
-default_flow = "0000"
+default_flow = "default"
 default_collection = "default"
 
 def unload_kg_core(url, user, id, flow):

trustgraph-cli/setup.py

@@ -34,7 +34,7 @@ setuptools.setup(
     python_requires='>=3.8',
     download_url = "https://github.com/trustgraph-ai/trustgraph/archive/refs/tags/v" + version + ".tar.gz",
     install_requires=[
-        "trustgraph-base>=0.23,<0.24",
+        "trustgraph-base>=1.0,<1.1",
         "requests",
         "pulsar-client",
         "aiohttp",

trustgraph-embeddings-hf/setup.py

@@ -34,8 +34,8 @@ setuptools.setup(
     python_requires='>=3.8',
     download_url = "https://github.com/trustgraph-ai/trustgraph/archive/refs/tags/v" + version + ".tar.gz",
     install_requires=[
-        "trustgraph-base>=0.23,<0.24",
+        "trustgraph-base>=1.0,<1.1",
-        "trustgraph-flow>=0.23,<0.24",
+        "trustgraph-flow>=1.0,<1.1",
         "torch",
         "urllib3",
         "transformers",

trustgraph-flow/scripts/text-completion-tgi

@@ -0,0 +1,6 @@
+#!/usr/bin/env python3
+
+from trustgraph.model.text_completion.tgi import run
+
+run()
+

trustgraph-flow/setup.py

@@ -34,7 +34,7 @@ setuptools.setup(
     python_requires='>=3.8',
     download_url = "https://github.com/trustgraph-ai/trustgraph/archive/refs/tags/v" + version + ".tar.gz",
     install_requires=[
-        "trustgraph-base>=0.23,<0.24",
+        "trustgraph-base>=1.0,<1.1",
         "aiohttp",
         "anthropic",
         "cassandra-driver",
@@ -42,7 +42,7 @@ setuptools.setup(
         "cryptography",
         "falkordb",
         "fastembed",
-        "google-generativeai",
+        "google-genai",
         "ibis",
         "jsonschema",
         "langchain",
@@ -117,6 +117,7 @@ setuptools.setup(
         "scripts/text-completion-mistral",
         "scripts/text-completion-ollama",
         "scripts/text-completion-openai",
+        "scripts/text-completion-tgi",
         "scripts/triples-query-cassandra",
         "scripts/triples-query-falkordb",
         "scripts/triples-query-memgraph",

trustgraph-flow/trustgraph/gateway/dispatch/librarian.py

@@ -65,8 +65,6 @@ class LibrarianRequestor(ServiceRequestor):
 
     def from_response(self, message):
 
-        print(message)
-
         response = {}
 
         if message.document_metadata:

trustgraph-flow/trustgraph/gateway/dispatch/mux.py

@@ -149,7 +149,6 @@ class Mux:
                 break
 
             try:
-                print(id, svc, request)
 
                 await self.start_request_task(
                     self.ws, id, flow, svc, request, workers

trustgraph-flow/trustgraph/gateway/dispatch/requestor.py

@@ -82,8 +82,6 @@ class ServiceRequestor:
 
                 resp, fin = self.from_response(resp)
 
-                print(resp, fin, flush=True)
-
                 if responder:
                     await responder(resp, fin)
 

trustgraph-flow/trustgraph/gateway/dispatch/streamer.py

@@ -76,8 +76,6 @@ class ServiceRequestor:
 
                     resp, fin = self.from_inbound(resp)
 
-                    print(resp, fin)
-
                     await self.receiver(resp, fin)
 
                 if fin: break

trustgraph-flow/trustgraph/gateway/dispatch/triples_query.py

@@ -48,7 +48,6 @@ class TriplesQueryRequestor(ServiceRequestor):
         )
 
     def from_response(self, message):
-        print(message)
         return {
             "response": serialize_subgraph(message.triples)
         }, True

trustgraph-flow/trustgraph/gateway/endpoint/constant_endpoint.py

@@ -47,10 +47,8 @@ class ConstantEndpoint:
 
             data = await request.json()
 
-            print(data)
-
             async def responder(x, fin):
-                print(x)
+                pass
 
             resp = await self.dispatcher.process(data, responder)
 

trustgraph-flow/trustgraph/gateway/endpoint/variable_endpoint.py

@@ -47,10 +47,8 @@ class VariableEndpoint:
 
             data = await request.json()
 
-            print(data)
-
             async def responder(x, fin):
-                print(x)
+                pass
 
             resp = await self.dispatcher.process(
                 data, responder, request.match_info

trustgraph-flow/trustgraph/model/text_completion/googleaistudio/llm.py

@@ -4,10 +4,18 @@ Simple LLM service, performs text prompt completion using GoogleAIStudio.
 Input is prompt, output is response.
 """
 
-import google.generativeai as genai
+#
-from google.generativeai.types import HarmCategory, HarmBlockThreshold
+# Using this SDK:
+#   https://googleapis.github.io/python-genai/genai.html#module-genai.client
+#
+# Seems to have simpler dependencies on the 'VertexAI' service, which
+# TrustGraph implements in the trustgraph-vertexai package.
+#
+
+from google import genai
+from google.genai import types
+from google.genai.types import HarmCategory, HarmBlockThreshold
 from google.api_core.exceptions import ResourceExhausted
-from prometheus_client import Histogram
 import os
 
 from .... exceptions import TooManyRequests
@@ -15,7 +23,7 @@ from .... base import LlmService, LlmResult
 
 default_ident = "text-completion"
 
-default_model = 'gemini-1.5-flash-002'
+default_model = 'gemini-2.0-flash-001'
 default_temperature = 0.0
 default_max_output = 8192
 default_api_key = os.getenv("GOOGLE_AI_STUDIO_KEY")
@@ -40,58 +48,56 @@ class Processor(LlmService):
             }
         )
 
-        genai.configure(api_key=api_key)
+        self.client = genai.Client(api_key=api_key)
         self.model = model
         self.temperature = temperature
         self.max_output = max_output
 
-        self.generation_config = {
-            "temperature": temperature,
-            "top_p": 1,
-            "top_k": 40,
-            "max_output_tokens": max_output,
-            "response_mime_type": "text/plain",
-        }
-
         block_level = HarmBlockThreshold.BLOCK_ONLY_HIGH
 
-        self.safety_settings={
+        self.safety_settings = [
-            HarmCategory.HARM_CATEGORY_HATE_SPEECH: block_level,
+            types.SafetySetting(
-            HarmCategory.HARM_CATEGORY_HARASSMENT: block_level,
+                category = HarmCategory.HARM_CATEGORY_HATE_SPEECH,
-            HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT: block_level,
+                threshold = block_level,
-            HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: block_level,
+            ),
+            types.SafetySetting(
+                category = HarmCategory.HARM_CATEGORY_HARASSMENT,
+                threshold = block_level,
+            ),
+            types.SafetySetting(
+                category = HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT,
+                threshold = block_level,
+            ),
+            types.SafetySetting(
+                category = HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
+                threshold = block_level,
+            ),
             # There is a documentation conflict on whether or not
             # CIVIC_INTEGRITY is a valid category
             # HarmCategory.HARM_CATEGORY_CIVIC_INTEGRITY: block_level,
-        }
+        ]
-
-        self.llm = genai.GenerativeModel(
-            model_name=model,
-            generation_config=self.generation_config,
-            safety_settings=self.safety_settings,
-            system_instruction="You are a helpful AI assistant.",
-        )
 
         print("Initialised", flush=True)
 
     async def generate_content(self, system, prompt):
 
-        # FIXME: There's a system prompt above.  Maybe if system changes,
+        generation_config = types.GenerateContentConfig(
-        # then reset self.llm?  It shouldn't do, because system prompt
+            temperature = self.temperature,
-        # is set system wide?
+            top_p = 1,
-
+            top_k = 40,
-        # Or... could keep different LLM structures for different system
+            max_output_tokens = self.max_output,
-        # prompts?
+            response_mime_type = "text/plain",
-
+            system_instruction = system,
-        prompt = system + "\n\n" + prompt
+            safety_settings = self.safety_settings,
+        )
 
         try:
 
-            chat_session = self.llm.start_chat(
+            response = self.client.models.generate_content(
-                history=[
+                model=self.model,
-                ]
+                config=generation_config,
+                contents=prompt,
             )
-            response = chat_session.send_message(prompt)
 
             resp = response.text
             inputtokens = int(response.usage_metadata.prompt_token_count)
@@ -158,3 +164,4 @@ class Processor(LlmService):
 def run():
     
     Processor.launch(default_ident, __doc__)
+

trustgraph-flow/trustgraph/model/text_completion/tgi/__init__.py

@@ -0,0 +1,3 @@
+
+from . llm import *
+

trustgraph-flow/trustgraph/model/text_completion/tgi/__main__.py

@@ -0,0 +1,7 @@
+#!/usr/bin/env python3
+
+from . llm import run
+
+if __name__ == '__main__':
+    run()
+

trustgraph-flow/trustgraph/model/text_completion/tgi/llm.py

@@ -0,0 +1,137 @@
+
+"""
+Simple LLM service, performs text prompt completion using HuggingFace TGI
+Input is prompt, output is response.
+"""
+
+import os
+import aiohttp
+
+from .... exceptions import TooManyRequests
+from .... base import LlmService, LlmResult
+
+default_ident = "text-completion"
+
+default_temperature = 0.0
+default_max_output = 2048
+default_base_url = os.getenv("TGI_BASE_URL")
+
+if default_base_url == "" or default_base_url is None:
+    default_base_url = "http://tgi-service:8899/v1"
+
+class Processor(LlmService):
+
+    def __init__(self, **params):
+    
+        base_url = params.get("url", default_base_url)
+        temperature = params.get("temperature", default_temperature)
+        max_output = params.get("max_output", default_max_output)
+
+        super(Processor, self).__init__(
+            **params | {
+                "temperature": temperature,
+                "max_output": max_output,
+                "url": base_url,
+            }
+        )
+
+        self.base_url = base_url
+        self.temperature = temperature
+        self.max_output = max_output
+
+        self.session = aiohttp.ClientSession()
+
+        print("Using TGI service at", base_url)
+
+        print("Initialised", flush=True)
+
+    async def generate_content(self, system, prompt):
+
+        headers = {
+            "Content-Type": "application/json",
+        }
+
+        request = {
+            "model": "tgi",
+            "messages": [
+                {
+                    "role": "system",
+                    "content": system,
+                },
+                {
+                    "role": "user",
+                    "content": prompt,
+                }
+            ],
+            "max_tokens": self.max_output,
+            "temperature": self.temperature,
+        }            
+
+        try:
+
+            url = f"{self.base_url}/chat/completions"
+
+            async with self.session.post(
+                    url,
+                    headers=headers,
+                    json=request,
+            ) as response:
+
+                if response.status != 200:
+                    raise RuntimeError("Bad status: " + str(response.status))
+
+                resp = await response.json()
+            
+            inputtokens = resp["usage"]["prompt_tokens"]
+            outputtokens = resp["usage"]["completion_tokens"]
+            ans = resp["choices"][0]["message"]["content"]
+            print(f"Input Tokens: {inputtokens}", flush=True)
+            print(f"Output Tokens: {outputtokens}", flush=True)
+            print(ans, flush=True)
+
+            resp = LlmResult(
+                text = ans,
+                in_token = inputtokens,
+                out_token = outputtokens,
+                model = "tgi",
+            )
+
+            return resp
+
+        # FIXME: Assuming TGI won't produce rate limits?
+
+        except Exception as e:
+
+            # Apart from rate limits, treat all exceptions as unrecoverable
+
+            print(f"Exception: {type(e)} {e}")
+            raise e
+
+    @staticmethod
+    def add_args(parser):
+
+        LlmService.add_args(parser)
+
+        parser.add_argument(
+            '-u', '--url',
+            default=default_base_url,
+            help=f'TGI service base URL (default: {default_base_url})'
+        )
+
+        parser.add_argument(
+            '-t', '--temperature',
+            type=float,
+            default=default_temperature,
+            help=f'LLM temperature parameter (default: {default_temperature})'
+        )
+
+        parser.add_argument(
+            '-x', '--max-output',
+            type=int,
+            default=default_max_output,
+            help=f'LLM max output tokens (default: {default_max_output})'
+        )
+
+def run():
+    
+    Processor.launch(default_ident, __doc__)

trustgraph-ocr/setup.py

@@ -34,7 +34,7 @@ setuptools.setup(
     python_requires='>=3.8',
     download_url = "https://github.com/trustgraph-ai/trustgraph/archive/refs/tags/v" + version + ".tar.gz",
     install_requires=[
-        "trustgraph-base>=0.23,<0.24",
+        "trustgraph-base>=1.0,<1.1",
         "pulsar-client",
         "prometheus-client",
         "boto3",

trustgraph-vertexai/setup.py

@@ -34,7 +34,7 @@ setuptools.setup(
     python_requires='>=3.8',
     download_url = "https://github.com/trustgraph-ai/trustgraph/archive/refs/tags/v" + version + ".tar.gz",
     install_requires=[
-        "trustgraph-base>=0.23,<0.24",
+        "trustgraph-base>=1.0,<1.1",
         "pulsar-client",
         "google-cloud-aiplatform",
         "prometheus-client",

trustgraph-vertexai/trustgraph/model/text_completion/vertexai/llm.py

@@ -4,13 +4,26 @@ Simple LLM service, performs text prompt completion using VertexAI on
 Google Cloud.   Input is prompt, output is response.
 """
 
+# 
+# Somewhat perplexed by the Google Cloud SDK choices.  We're going off this
+# one, which uses the google-cloud-aiplatform library:
+#   https://cloud.google.com/python/docs/reference/vertexai/1.94.0
+# It seems it is possible to invoke VertexAI from the google-genai
+# SDK too:
+#   https://googleapis.github.io/python-genai/genai.html#module-genai.client
+# That would make this code look very much like the GoogleAIStudio
+# code.  And maybe not reliant on the google-cloud-aiplatform library?
+#
+# This module's imports bring in a lot of libraries.
+
 from google.oauth2 import service_account
 import google
 import vertexai
 
-from vertexai.preview.generative_models import (
+# Why is preview here?
+from vertexai.generative_models import (
     Content, FunctionDeclaration, GenerativeModel, GenerationConfig,
-    HarmCategory, HarmBlockThreshold, Part, Tool,
+    HarmCategory, HarmBlockThreshold, Part, Tool, SafetySetting,
 )
 
 from .... exceptions import TooManyRequests
@@ -18,7 +31,7 @@ from .... base import LlmService, LlmResult
 
 default_ident = "text-completion"
 
-default_model = 'gemini-1.0-pro-001'
+default_model = 'gemini-2.0-flash-001'
 default_region = 'us-central1'
 default_temperature = 0.0
 default_max_output = 8192
@@ -59,12 +72,24 @@ class Processor(LlmService):
         block_level = HarmBlockThreshold.BLOCK_ONLY_HIGH
         #     block_level = HarmBlockThreshold.BLOCK_NONE
 
-        self.safety_settings = {
+        self.safety_settings = [
-            HarmCategory.HARM_CATEGORY_HARASSMENT: block_level,
+            SafetySetting(
-            HarmCategory.HARM_CATEGORY_HATE_SPEECH: block_level,
+                category = HarmCategory.HARM_CATEGORY_HARASSMENT,
-            HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT: block_level,
+                threshold = block_level,
-            HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT: block_level,
+            ),
-        }
+            SafetySetting(
+                category = HarmCategory.HARM_CATEGORY_HATE_SPEECH,
+                threshold = block_level,
+            ),
+            SafetySetting(
+                category = HarmCategory.HARM_CATEGORY_SEXUALLY_EXPLICIT,
+                threshold = block_level,
+            ),
+            SafetySetting(
+                category = HarmCategory.HARM_CATEGORY_DANGEROUS_CONTENT,
+                threshold = block_level,
+            ),
+        ]
 
         print("Initialise VertexAI...", flush=True)
 
@@ -101,8 +126,8 @@ class Processor(LlmService):
             prompt = system + "\n\n" + prompt
 
             response = self.llm.generate_content(
-                prompt, generation_config=self.generation_config,
+                prompt, generation_config = self.generation_config,
-                safety_settings=self.safety_settings
+                safety_settings = self.safety_settings,
             )
 
             resp = LlmResult(

trustgraph/setup.py

@@ -34,12 +34,12 @@ setuptools.setup(
     python_requires='>=3.8',
     download_url = "https://github.com/trustgraph-ai/trustgraph/archive/refs/tags/v" + version + ".tar.gz",
     install_requires=[
-        "trustgraph-base>=0.23,<0.24",
+        "trustgraph-base>=1.0,<1.1",
-        "trustgraph-bedrock>=0.23,<0.24",
+        "trustgraph-bedrock>=1.0,<1.1",
-        "trustgraph-cli>=0.23,<0.24",
+        "trustgraph-cli>=1.0,<1.1",
-        "trustgraph-embeddings-hf>=0.23,<0.24",
+        "trustgraph-embeddings-hf>=1.0,<1.1",
-        "trustgraph-flow>=0.23,<0.24",
+        "trustgraph-flow>=1.0,<1.1",
-        "trustgraph-vertexai>=0.23,<0.24",
+        "trustgraph-vertexai>=1.0,<1.1",
     ],
     scripts=[
     ]